aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
-rw-r--r--src/mesa/main/drawpix.c21
-rw-r--r--src/mesa/main/teximage.c178
2 files changed, 152 insertions, 47 deletions
diff --git a/src/mesa/main/drawpix.c b/src/mesa/main/drawpix.c
index 562db485b53..93dd66a0429 100644
--- a/src/mesa/main/drawpix.c
+++ b/src/mesa/main/drawpix.c
@@ -1,4 +1,4 @@
-/* $Id: drawpix.c,v 1.30 2000/08/23 14:32:06 brianp Exp $ */
+/* $Id: drawpix.c,v 1.31 2000/08/30 18:22:28 brianp Exp $ */
/*
* Mesa 3-D graphics library
@@ -746,17 +746,6 @@ draw_rgba_pixels( GLcontext *ctx, GLint x, GLint y,
* we'll proceed with the rest of the transfer operations and
* rasterize the image.
*/
- const GLuint preConvTransferOps =
- IMAGE_SCALE_BIAS_BIT |
- IMAGE_SHIFT_OFFSET_BIT |
- IMAGE_MAP_COLOR_BIT |
- IMAGE_COLOR_TABLE_BIT;
- const GLuint postConvTransferOps =
- IMAGE_POST_CONVOLUTION_COLOR_TABLE_BIT |
- IMAGE_COLOR_MATRIX_BIT |
- IMAGE_POST_COLOR_MATRIX_COLOR_TABLE_BIT |
- IMAGE_HISTOGRAM_BIT |
- IMAGE_MIN_MAX_BIT;
GLint row;
GLfloat *dest, *tmpImage;
@@ -778,9 +767,9 @@ draw_rgba_pixels( GLcontext *ctx, GLint x, GLint y,
const GLvoid *source = _mesa_image_address(unpack,
pixels, width, height, format, type, 0, row, 0);
_mesa_unpack_float_color_span(ctx, width, GL_RGBA, (void *) dest,
- format, type, source, unpack,
- transferOps & preConvTransferOps,
- GL_FALSE);
+ format, type, source, unpack,
+ transferOps & IMAGE_PRE_CONVOLUTION_BITS,
+ GL_FALSE);
dest += width * 4;
}
@@ -799,7 +788,7 @@ draw_rgba_pixels( GLcontext *ctx, GLint x, GLint y,
pixels = convImage;
format = GL_RGBA;
type = GL_FLOAT;
- transferOps &= postConvTransferOps;
+ transferOps &= IMAGE_POST_CONVOLUTION_BITS;
}
/*
diff --git a/src/mesa/main/teximage.c b/src/mesa/main/teximage.c
index db4b51f7316..e643bfd60a3 100644
--- a/src/mesa/main/teximage.c
+++ b/src/mesa/main/teximage.c
@@ -1,4 +1,4 @@
-/* $Id: teximage.c,v 1.41 2000/08/29 23:31:23 brianp Exp $ */
+/* $Id: teximage.c,v 1.42 2000/08/30 18:22:28 brianp Exp $ */
/*
* Mesa 3-D graphics library
@@ -30,6 +30,7 @@
#else
#include "glheader.h"
#include "context.h"
+#include "convolve.h"
#include "image.h"
#include "mem.h"
#include "mmath.h"
@@ -636,6 +637,39 @@ _mesa_select_tex_image(GLcontext *ctx, const struct gl_texture_unit *texUnit,
+/*
+ * Calling glTexImage and related functions when convolution is enabled
+ * with GL_REDUCE border mode causes some complications.
+ * The incoming image must be extra large so that the post-convolution
+ * image size is reduced to a power of two size (plus 2 * border).
+ * This function adjusts a texture width and height accordingly if
+ * convolution with GL_REDUCE is enabled.
+ */
+static void
+adjust_texture_size_for_convolution(const GLcontext *ctx, GLuint dimensions,
+ GLsizei *width, GLsizei *height)
+{
+ if (ctx->Pixel.Convolution1DEnabled
+ && dimensions == 1
+ && ctx->Pixel.ConvolutionBorderMode[0] == GL_REDUCE) {
+ *width = *width - (MAX2(ctx->Convolution1D.Width, 1) - 1);
+ }
+ else if (ctx->Pixel.Convolution2DEnabled
+ && dimensions > 1
+ && ctx->Pixel.ConvolutionBorderMode[1] == GL_REDUCE) {
+ *width = *width - (MAX2(ctx->Convolution2D.Width, 1) - 1);
+ *height = *height - (MAX2(ctx->Convolution2D.Height, 1) - 1);
+ }
+ else if (ctx->Pixel.Separable2DEnabled
+ && dimensions > 1
+ && ctx->Pixel.ConvolutionBorderMode[2] == GL_REDUCE) {
+ *width = *width - (MAX2(ctx->Separable2D.Width, 1) - 1);
+ *height = *height - (MAX2(ctx->Separable2D.Height, 1) - 1);
+ }
+}
+
+
+
/* Need this to prevent an out-of-bounds memory access when using
* X86 optimized code.
*/
@@ -653,10 +687,10 @@ _mesa_select_tex_image(GLcontext *ctx, const struct gl_texture_unit *texUnit,
* NOTE: All texture image parameters should have already been error checked.
*/
static void
-make_texture_image( GLcontext *ctx,
+make_texture_image( GLcontext *ctx, GLuint dimensions,
struct gl_texture_image *texImage,
GLenum srcFormat, GLenum srcType, const GLvoid *pixels,
- const struct gl_pixelstore_attrib *unpacking)
+ const struct gl_pixelstore_attrib *srcPacking)
{
GLint components, numPixels;
GLint internalFormat, width, height, depth, border;
@@ -665,7 +699,7 @@ make_texture_image( GLcontext *ctx,
ASSERT(texImage);
ASSERT(!texImage->Data);
ASSERT(pixels);
- ASSERT(unpacking);
+ ASSERT(srcPacking);
internalFormat = texImage->IntFormat;
width = texImage->Width;
@@ -678,8 +712,6 @@ make_texture_image( GLcontext *ctx,
ASSERT(height > 0);
ASSERT(depth > 0);
ASSERT(border == 0 || border == 1);
- ASSERT(pixels);
- ASSERT(unpacking);
ASSERT(components);
numPixels = width * height * depth;
@@ -710,8 +742,8 @@ make_texture_image( GLcontext *ctx,
* GL_LUMINANCE_ALPHA, etc. texture formats.
*/
const GLubyte *src = (const GLubyte *) _mesa_image_address(
- unpacking, pixels, width, height, srcFormat, srcType, 0, 0, 0);
- const GLint srcStride = _mesa_image_row_stride(unpacking, width,
+ srcPacking, pixels, width, height, srcFormat, srcType, 0, 0, 0);
+ const GLint srcStride = _mesa_image_row_stride(srcPacking, width,
srcFormat, srcType);
GLubyte *dst = texImage->Data;
GLint dstBytesPerRow = width * components * sizeof(GLubyte);
@@ -731,8 +763,8 @@ make_texture_image( GLcontext *ctx,
else if (srcFormat == GL_RGBA && internalFormat == GL_RGB) {
/* commonly used by Quake */
const GLubyte *src = (const GLubyte *) _mesa_image_address(
- unpacking, pixels, width, height, srcFormat, srcType, 0, 0, 0);
- const GLint srcStride = _mesa_image_row_stride(unpacking, width,
+ srcPacking, pixels, width, height, srcFormat, srcType, 0, 0, 0);
+ const GLint srcStride = _mesa_image_row_stride(srcPacking, width,
srcFormat, srcType);
GLubyte *dst = texImage->Data;
GLint i, j;
@@ -762,10 +794,10 @@ make_texture_image( GLcontext *ctx,
GLint img, row;
for (img = 0; img < depth; img++) {
for (row = 0; row < height; row++) {
- const GLvoid *source = _mesa_image_address(unpacking,
+ const GLvoid *srcAddr = _mesa_image_address(srcPacking,
pixels, width, height, srcFormat, srcType, img, row, 0);
_mesa_unpack_index_span(ctx, width, dstType, dest,
- srcType, source, unpacking,
+ srcType, srcAddr, srcPacking,
ctx->ImageTransferState);
dest += destBytesPerRow;
}
@@ -777,15 +809,86 @@ make_texture_image( GLcontext *ctx,
const GLenum dstFormat = texImage->Format;
GLubyte *dest = texImage->Data;
GLint img, row;
- /* XXX convolution */
- for (img = 0; img < depth; img++) {
- for (row = 0; row < height; row++) {
- const GLvoid *source = _mesa_image_address(unpacking,
- pixels, width, height, srcFormat, srcType, img, row, 0);
- _mesa_unpack_ubyte_color_span(ctx, width, dstFormat, dest,
- srcFormat, srcType, source,
- unpacking, ctx->ImageTransferState);
- dest += destBytesPerRow;
+ GLint w = width, h = height;
+
+ if ((dimensions == 1 && ctx->Pixel.Convolution1DEnabled) ||
+ (dimensions >= 2 &&
+ (ctx->Pixel.Convolution2DEnabled || ctx->Pixel.Separable2DEnabled)
+ )) {
+ GLfloat *tmpImage, *convImage;
+ tmpImage = (GLfloat *) MALLOC(width * height * sizeof(GLfloat));
+ if (!tmpImage) {
+ gl_error(ctx, GL_OUT_OF_MEMORY, "glTexImage");
+ return;
+ }
+ convImage = (GLfloat *) MALLOC(width * height * sizeof(GLfloat));
+ if (!convImage) {
+ gl_error(ctx, GL_OUT_OF_MEMORY, "glTexImage");
+ FREE(tmpImage);
+ return;
+ }
+
+ for (img = 0; img < depth; img++) {
+ const GLfloat *srcf;
+ GLfloat *dstf = tmpImage;
+ /* unpack and do transfer ops up to convolution */
+ for (row = 0; row < height; row++) {
+ const GLvoid *srcAddr = _mesa_image_address(srcPacking,
+ pixels, width, height, srcFormat, srcType, img, row, 0);
+ _mesa_unpack_float_color_span(ctx, width, GL_RGBA, dstf,
+ srcFormat, srcType, srcAddr, srcPacking,
+ ctx->ImageTransferState & IMAGE_PRE_CONVOLUTION_BITS,
+ GL_TRUE);
+ dstf += width * 4;
+ }
+
+ /* convolve */
+ if (dimensions == 1) {
+ if (ctx->Pixel.Convolution1DEnabled) {
+ _mesa_convolve_1d_image(ctx, &w, tmpImage, convImage);
+ }
+ }
+ else {
+ if (ctx->Pixel.Convolution2DEnabled) {
+ _mesa_convolve_2d_image(ctx, &w, &h, tmpImage, convImage);
+ }
+ else {
+ ASSERT(ctx->Pixel.Separable2DEnabled);
+ _mesa_convolve_sep_image(ctx, &w, &h, tmpImage, convImage);
+ }
+ }
+
+ /* transfer ops after convolution */
+ srcf = convImage;
+ for (row = 0; row < h; row++) {
+ GLvoid *dest;
+ dest = _mesa_image_address(&_mesa_native_packing, pixels,
+ w, h, GL_RGBA, GL_UNSIGNED_BYTE,
+ 0, row, 0);
+ _mesa_pack_float_rgba_span(ctx, w,
+ (const GLfloat (*)[4]) srcf,
+ dstFormat, GL_UNSIGNED_BYTE,
+ dest, &_mesa_native_packing,
+ ctx->ImageTransferState
+ & IMAGE_POST_CONVOLUTION_BITS);
+ srcf += w * 4;
+ }
+ }
+
+ FREE(convImage);
+ FREE(tmpImage);
+ }
+ else {
+ /* no convolution */
+ for (img = 0; img < depth; img++) {
+ for (row = 0; row < height; row++) {
+ const GLvoid *srcAddr = _mesa_image_address(srcPacking,
+ pixels, width, height, srcFormat, srcType, img, row, 0);
+ _mesa_unpack_ubyte_color_span(ctx, width, dstFormat, dest,
+ srcFormat, srcType, srcAddr, srcPacking,
+ ctx->ImageTransferState);
+ dest += destBytesPerRow;
+ }
}
}
}
@@ -1403,7 +1506,8 @@ get_specific_compressed_tex_format(GLcontext *ctx,
internalFormat = GL_RGBA;
break;
default:
- gl_problem(ctx, "unexpected format in get_specific_compressed_tex_format");
+ /* silence compiler warning */
+ ;
}
return internalFormat;
}
@@ -1418,9 +1522,13 @@ _mesa_TexImage1D( GLenum target, GLint level, GLint internalFormat,
GLsizei width, GLint border, GLenum format,
GLenum type, const GLvoid *pixels )
{
+ GLsizei postConvWidth;
GET_CURRENT_CONTEXT(ctx);
ASSERT_OUTSIDE_BEGIN_END_AND_FLUSH(ctx, "glTexImage1D");
+ postConvWidth = width;
+ adjust_texture_size_for_convolution(ctx, 1, &postConvWidth, NULL);
+
if (target==GL_TEXTURE_1D) {
struct gl_texture_unit *texUnit;
struct gl_texture_object *texObj;
@@ -1440,7 +1548,7 @@ _mesa_TexImage1D( GLenum target, GLint level, GLint internalFormat,
}
if (texture_error_check(ctx, target, level, internalFormat,
- format, type, 1, width, 1, 1, border)) {
+ format, type, 1, postConvWidth, 1, 1, border)) {
return; /* error in texture image was detected */
}
@@ -1462,7 +1570,7 @@ _mesa_TexImage1D( GLenum target, GLint level, GLint internalFormat,
}
/* setup the teximage struct's fields */
- init_texture_image(ctx, texImage, width, 1, 1, border, internalFormat);
+ init_texture_image(ctx, texImage, postConvWidth, 1, 1, border, internalFormat);
if (ctx->ImageTransferState == UPDATE_IMAGE_TRANSFER_STATE)
_mesa_update_image_transfer_state(ctx);
@@ -1479,7 +1587,7 @@ _mesa_TexImage1D( GLenum target, GLint level, GLint internalFormat,
}
if (retain || !success) {
/* make internal copy of the texture image */
- make_texture_image(ctx, texImage, format, type,
+ make_texture_image(ctx, 1, texImage, format, type,
pixels, &ctx->Unpack);
if (!success && ctx->Driver.TexImage1D) {
/* let device driver try to use unpacked image */
@@ -1512,7 +1620,7 @@ _mesa_TexImage1D( GLenum target, GLint level, GLint internalFormat,
else if (target==GL_PROXY_TEXTURE_1D) {
/* Proxy texture: check for errors and update proxy state */
if (texture_error_check(ctx, target, level, internalFormat,
- format, type, 1, width, 1, 1, border)) {
+ format, type, 1, postConvWidth, 1, 1, border)) {
/* if error, clear all proxy texture image parameters */
if (level>=0 && level<ctx->Const.MaxTextureLevels) {
MEMSET( ctx->Texture.Proxy1D->Image[level], 0,
@@ -1538,9 +1646,14 @@ _mesa_TexImage2D( GLenum target, GLint level, GLint internalFormat,
GLenum format, GLenum type,
const GLvoid *pixels )
{
+ GLsizei postConvWidth, postConvHeight;
GET_CURRENT_CONTEXT(ctx);
ASSERT_OUTSIDE_BEGIN_END_AND_FLUSH(ctx, "glTexImage2D");
+ postConvWidth = width;
+ postConvHeight = height;
+ adjust_texture_size_for_convolution(ctx, 2, &postConvWidth,&postConvHeight);
+
if (target==GL_TEXTURE_2D ||
(ctx->Extensions.HaveTextureCubeMap &&
target >= GL_TEXTURE_CUBE_MAP_POSITIVE_X_ARB &&
@@ -1563,7 +1676,8 @@ _mesa_TexImage2D( GLenum target, GLint level, GLint internalFormat,
}
if (texture_error_check(ctx, target, level, internalFormat,
- format, type, 2, width, height, 1, border)) {
+ format, type, 2, postConvWidth, postConvHeight,
+ 1, border)) {
return; /* error in texture image was detected */
}
@@ -1586,7 +1700,7 @@ _mesa_TexImage2D( GLenum target, GLint level, GLint internalFormat,
}
/* setup the teximage struct's fields */
- init_texture_image(ctx, texImage, width, height,
+ init_texture_image(ctx, texImage, postConvWidth, postConvHeight,
1, border, internalFormat);
if (ctx->ImageTransferState == UPDATE_IMAGE_TRANSFER_STATE)
@@ -1604,7 +1718,7 @@ _mesa_TexImage2D( GLenum target, GLint level, GLint internalFormat,
}
if (retain || !success) {
/* make internal copy of the texture image */
- make_texture_image(ctx, texImage, format, type,
+ make_texture_image(ctx, 2, texImage, format, type,
pixels, &ctx->Unpack);
if (!success && ctx->Driver.TexImage2D) {
/* let device driver try to use unpacked image */
@@ -1646,7 +1760,7 @@ _mesa_TexImage2D( GLenum target, GLint level, GLint internalFormat,
else if (target==GL_PROXY_TEXTURE_2D) {
/* Proxy texture: check for errors and update proxy state */
if (texture_error_check(ctx, target, level, internalFormat,
- format, type, 2, width, height, 1, border)) {
+ format, type, 2, postConvWidth, height, 1, border)) {
/* if error, clear all proxy texture image parameters */
if (level>=0 && level<ctx->Const.MaxTextureLevels) {
MEMSET( ctx->Texture.Proxy2D->Image[level], 0,
@@ -1740,7 +1854,7 @@ _mesa_TexImage3D( GLenum target, GLint level, GLint internalFormat,
}
if (retain || !success) {
/* make internal copy of the texture image */
- make_texture_image(ctx, texImage, format, type,
+ make_texture_image(ctx, 3, texImage, format, type,
pixels, &ctx->Unpack);
if (!success && ctx->Driver.TexImage3D) {
/* let device driver try to use unpacked image */
@@ -1997,6 +2111,7 @@ _mesa_GetTexImage( GLenum target, GLint level, GLenum format,
assert(dest);
if (texImage->Format == GL_RGBA) {
const GLubyte *src = texImage->Data + row * width * 4 * sizeof(GLubyte);
+ /* XXX convolution */
_mesa_pack_rgba_span( ctx, width, (CONST GLubyte (*)[4]) src,
format, type, dest, &ctx->Pack,
ctx->ImageTransferState );
@@ -2062,6 +2177,7 @@ _mesa_GetTexImage( GLenum target, GLint level, GLenum format,
default:
gl_problem( ctx, "bad format in gl_GetTexImage" );
}
+ /* XXX convolution */
_mesa_pack_rgba_span( ctx, width, (const GLubyte (*)[4])rgba,
format, type, dest, &ctx->Pack,
ctx->ImageTransferState );